The Comparative Level of GDP per Capita in Canada and the United States: A Decomposition into Labour Productivity and Work Intensity Differences

Transcription

1 Catalogue no XIE No. 008 ISSN: O ISBN: Research Paper The Canadian Productivity Review The Comparative Level of GDP per Capita in Canada and the United States: A Decomposition into Labour Productivity and Work Intensity Differences by Jean-Pierre Maynard Telephone:

2 How to obtain more information Specific inquiries about this product and related statistics or services should be directed to the Media Hotline, Communications and Library Services Division, Statistics Canada, Ottawa, Ontario K1A 0T6 (telephone: ). For information on the wide range of data available from Statistics Canada, you can contact us by calling one of our tollfree numbers. You can also contact us by or by visiting our website at National inquiries line National telecommunications device for the hearing impaired Depository Services Program inquiries Fax line for Depository Services Program inquiries Website Information to access the product This product, catalogue no XIE, is available for free in electronic format. To obtain a single issue, visit our website at and select Publications. Standards of service to the public Statistics Canada is committed to serving its clients in a prompt, reliable and courteous manner and in the official language of their choice. To this end, it has developed standards of service that its employees observe in serving its clients. To obtain a copy of these service standards, please contact Statistics Canada toll free at The service standards are also published on under About us > Providing services to Canadians. The Canadian Productivity Review The Canadian Productivity Review is a series of applied studies that address issues involving the measurement, explanation, and improvement of productivity. Themes covered in the review include, but are not limited to, economic performance, capital formation, labour, prices, environment, trade, and efficiency at both national and provincial levels. The Review publishes empirical research, at different levels of aggregation, based on growth accounting, econometrics, index numbers and mathematical programming. The empirical research illustrates the application of theory and techniques to relevant public policy issues. The primary distribution medium for this series is the Internet. These studies can be downloaded from the Internet at for free. All papers in The Canadian Productivity Review go through institutional review to ensure that they conform to Statistics Canada s mandate as a government statistical agency and adhere to generally accepted standards of good professional practice. The papers in the series often include results derived from multivariate analysis or other statistical techniques. It should be recognized that the results of these analyses are subject to uncertainty in the reported estimates. The level of uncertainty will depend on several factors: the nature of the functional form used in the multivariate analysis; the type of econometric technique employed; the appropriateness of the statistical assumptions embedded in the model or technique; the comprehensiveness of the variables included in the analysis; and the accuracy of the data that are utilized. The peer group review process is meant to ensure that the papers in the series have followed accepted standards to minimize problems in each of these areas.

3 Statistics Canada The Comparative Level of GDP per Capita in Canada and the United States: A Decomposition into Labour Productivity and Work Intensity Differences Jean-Pierre Maynard Published by authority of the Minister responsible for Statistics Canada Minister of Industry, 2007 All rights reserved. The content of this electronic publication may be reproduced, in whole or in part, and by any means, without further permission from Statistics Canada, subject to the following conditions: that it be done solely for the purposes of private study, research, criticism, review or newspaper summary, and/or for non-commercial purposes; and that Statistics Canada be fully acknowledged as follows: Source (or Adapted from, if appropriate): Statistics Canada, year of publication, name of product, catalogue number, volume and issue numbers, reference period and page(s). Otherwise, no part of this publication may be reproduced, stored in a retrieval system or transmitted in any form, by any means electronic, mechanical or photocopy or for any purposes without prior written permission of Licensing Services, Client Services Division, Statistics Canada, Ottawa, Ontario, Canada K1A 0T6. March 2007 Catalogue no XIE, no. 008 Frequency: Occasional ISSN ISBN Ottawa La version française de cette publication est disponible (n o XIF au catalogue, n o 008). Note of appreciation Canada owes the success of its statistical system to a long-standing partnership between Statistics Canada, the citizens of Canada, its businesses, governments and other institutions. Accurate and timely statistical information could not be produced without their continued cooperation and goodwill.

4 Acknowledgement The author would like to thank John Baldwin, Tarek Harchaoui and Mustapha Kaci for their invaluable help with the presentation and content of the various drafts that led to this final version. He also wishes to thank Don Drummond, Graham Rose and Gloria Wong for their relevant comments, as well as Mike Harper and Phyllis Otto from the Bureau of Labor Statistics for the many clarifications provided about U.S. labour statistics. This third article on the project comparing Canada United States productivity levels initiated in fall 2003 by the Canadian Productivity Accounts would never have seen the light of day without the outstanding work of a team of analysts composed of Marc Tanguay, Jin Lee, Fanny Wong and Sean Burrows. However, despite the involvement of all these people, the author remains wholly responsible for any error or omission in this study. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

5 Table of Contents Abstract... 6 Executive summary Introduction Comparison framework Estimation of labour input for comparisons of relative levels of labour productivity in Canada and the United States Criteria Concept Coverage Accuracy or quality of estimates Corroboration Sources of labour inputs Enterprise surveys Household surveys Example: Differences between household surveys and enterprise surveys Overall reliability of average hours worked derived from household surveys Estimating total hours worked Jobs Jobs versus employment Hours worked per job Measurement of population Framework for reconciliation between alternative measures Canada/U.S. differences Background Findings and implications Alternative M2 (using unadjusted data from the household labour force surveys) Alternative M3 (using U.S. data from the Bureau of Labor Statistics productivity growth programs) Conclusion Appendix 1 Comparison of hours between the Labour Force Survey and the Survey of Employment, Payrolls and Hours Appendix 2 Questions on hours from the Labour Force Survey and the Current Population Survey Appendix 3 Glossary References The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

6 Abstract This study is the third in a series related to the project launched in the fall of 2003 by the Canadian Productivity Accounts of Statistics Canada in order to compare productivity levels between Canada and the United States. The study s purpose is to examine the comparability of the components of the labour market of these two countries that serve as the sources of the differences in the gross domestic product (GDP) per capita between them. This study can be subdivided into three sections. The first section develops and illustrates the conceptual and methodological framework required to make Canada United States estimates of labour and population comparable in terms of level. The second section presents revisions and an update to 2005 of the GDP per capita differences and its components, which were presented for the first time in the study by Baldwin, Maynard and Wong (2005), which covered the period from 1994 to 2002, at the time. Lastly, using the year 2000 as an example, this study tries to quantify the statistical error that arises from using inadequate statistics or statistics not designed for this type of international comparison. This exercise reveals that the comparability of data on hours worked per job is especially crucial to identifying the origin of the differences in GDP per capita between labour productivity and hours worked per capita. The worst error involves comparing hours worked estimated from an employer survey with those obtained from a household survey. This type of comparison between Canada and the United States results in assigning an estimated 72% of the difference in GDP per capita to labour productivity when, in reality, it counted for barely 36% in The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

7 Executive summary This study is the third in a series related to the project launched in 2003 by Statistics Canada s Canadian Productivity Accounts (CPA) to compare productivity levels between Canada and the United States. In 2005, Statistics Canada s Canadian Productivity Accounts released two studies that, for the first time, examined the comparability of labour productivity levels between Canada and the United States. 1 Previously, Statistics Canada limited comparisons to productivity growth rates. Using analogous sources, concepts and methods to obtain the most comparable measure possible of productivity levels, these new studies found that the Canada U.S. productivity level difference was lower than normally described. Neither the Canadian nor the American data used to measure work intensity for this project are the same as those used by many who have conducted Canada/US comparisons of the level of labour productivity. Other studies have used data that were assumed to be comparable such as data from the Labour Force Survey (LFS) in Canada and those from the equivalent American survey, the Current Population Survey (CPS) but which are not. This third study focuses in more depth on the construction of the number of hours worked developed for this project and on the choice of estimates of jobs and population. It describes the reasons that the work intensity measures used in our Canada/U.S. project are superior to alternatives that are readily available but non comparable and therefore inappropriate for studies of Canada/U.S. comparisons of the level of productivity. This study answers the following questions: 1) What are the reasons for the choice of data to measure the number of hours worked? 2) Why are the estimates of the number of hours worked developed for this study the most appropriate for comparing levels of work intensity and hours worked per job between Canada and the United States? 3) What are the problems with traditional data sources that make them inappropriate for comparisons of levels? 4) What is the degree of error that is made if a study relies on alternate but easily accessible labour force sources to compare levels of productivity and work intensity between Canada and the United States? Using a framework to analyze differences in real GDP per capita that decomposes the latter into differences in labour productivity and work intensity, this study quantifies the errors committed when alternate, easily accessible but non comparable data sources are used in order to compare the sources of differences in GDP per capita between Canada and the United States. 1. Baldwin et al., 2005; Baldwin, Maynard and Wong, The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

8 Table comparing alternatives for decomposing the Canada U.S. difference in gross domestic product (GDP) per capita between labour productivity and work intensity and its sources % difference in relation to the United States (2000) Source data GDP per capita Labour productivity Work intensity Hours worked per capita Hours worked per job Work intensity components Job / population aged 15 years and over Population aged 15 years and over to population 1. CPA / BLS-PA LFS / CPS CPA project Notes: Differences are expressed in this study in logarithms to preserve their additivity. The three rows make use of different source data. Measure #1 compares official data for the economy as a whole that are used to measure labour productivity growth in the two countries. CPA is the acronym for the Canadian Productivity Accounts, while BLS-PA stands for Bureau of Labor Statistics - Productivity Accounts. In measure #2, LFS refers to Canada s Labour Force Survey and CPS stands for Current Population Survey, the American equivalent. Measure #3 presents results derived from the project to compare productivity levels conducted by the Canadian Productivity Accounts (CPA). This table presents the results of the decomposition of the Canada U.S. difference in GDP per capita for the year 2000 using two inappropriate measures that have been occasionally used for Canada/U.S. comparisons. The first inappropriate measure (line 1) uses estimates of labour input developed by the productivity program of each country to measure the growth in labour productivity. Note that the primary objective of these programs is to estimate productivity growth and not the level of productivity relative to other countries. The second measure (line 2) uses data coming from the monthly household surveys of the two countries the Labour Force Survey (LFS) in Canada and the Current Population Survey (CPS) in the United States. The third measure (line 3) makes use of the data on labour inputs generated in the Statistics Canada project that developed comparable data to be used to estimate the relative level of Canada U.S. productivity. Using the year 2000 as an example and the same GDP per capita measure for the three sources of components, this study shows the crucial importance of using comparable measures, which were developed by harmonizing concepts and coverage and by adjusting data to consider differences in collection methods and in data accuracy. The appropriate comparison (line 3) that uses comparable data shows that labour productivity contributes much less to GDP per capita differences than the two inappropriate techniques. Measure #1: Problem with hours per job The first inappropriate measure uses the levels of the number of hours worked and the number of jobs derived from the official measures used to estimate labour productivity growth in both countries. Using this comparison, 70% of the 20% gap in gross domestic product (GDP) per capita in favour of the United States in 2000 can be attributed to Canada s weaker level of labour productivity. The correct measure (line 3) indicates that only 35% of the gap is due to lower labour productivity. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

9 In general, both countries produce detailed estimates of the number of hours worked by estimating the number of jobs and the annual number of hours worked per job. The number of hours worked is obtained by multiplying these two elements. The Canadian Productivity Accounts rely mainly on a household survey, the Labour Force Survey (LFS), to estimate employment; in the United States, the starting point for constructing these same estimates is an employer survey, the Current Employment Statistics (CES). Given that this survey has only partial coverage (does not cover, for example, farms and self-employed workers), the Current Population Survey (CPS) estimates are used to complete the coverage. Our assessment is, that based on conceptual, coverage and accuracy criteria, these two measures of employment are appropriate for comparing employment levels between the two countries. The problem with measure #1 arises because the estimates of hours per job are derived from different types of surveys that in each country yield quite different estimates of hours worked per job. The Canadian Productivity Accounts rely on hours actually worked collected from a household survey the LFS; on the other hand, the Bureau of Labor Statistics (BLS) instead uses the hours paid collected from its survey of employers. Although the estimates of hours paid are then converted by the BLS into hours worked by excluding hours of paid leave (vacation, holidays, sick, etc.), these two approaches produce results that are not comparable because household and employer surveys produce estimates that differ in a systematic way. As part of this project, the Canadian Productivity Accounts conducted a comparison of the estimate of hours worked per job using household and employer surveys in each country. The results from similar surveys were compared across countries (household survey in Canada to household survey in the United States; employer survey in Canada to employer survey in the United States). The comparison for household surveys made use of a similar methodology to adjust these data for the bias associated with household surveys that do not take into account statutory holidays when extrapolating data from a survey reference week to other weeks in the month. The Canadian Survey of Employment, Payrolls and Hours was used (a survey of employers that gathers data on hours paid or on the regular hours of workers) to estimate the Canadian hours worked following the BLS method that uses an employer survey. This result was then compared to the average hours derived by the BLS from their employer survey. Each comparison indicated that Americans worked on average one hundred hours more than their Canadian counterparts, annually. But in both countries, the estimate derived from the employer survey was always below the estimate derived from the household survey. The hours worked per job estimated from the data collected from employers produced a level about 8% lower than the level derived from the household surveys after adjustment for statutory holiday bias. This demonstrates the direction and size of the error that occurs when a household survey in Canada is compared to an employer survey in the United States, as is done for measure #1. There still remains the issue of whether hours worked for comparisons of levels should be estimated from household or employer surveys. Various studies conducted in several countries, including Canada and the United States, have compared the estimates of hours worked collected The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

10 from households using a daily survey of time use in theory the best approach for collecting this information to the estimates derived both from employer and labour force surveys. The estimates derived from the time-use surveys suggest that labour force surveys provide the most accurate estimates of hours per job. Therefore, these are the estimates that have been adopted in our Canada/U.S. comparison. This first example demonstrates that the source of the data on hours worked per job is especially important in order to attribute the origin of GDP per capita differences to labour productivity or to hours worked per capita. Comparing hours worked estimated from a survey of employers to those obtained from a household survey has the potential to overestimate the impact of productivity differences on GDP per capita differences between Canada and the United States. Measure #2: Sources of labour intensity The second inappropriate measure (line 2) compares the levels of the number of hours worked, the number of persons employed and the civilian population of working age outside institutions obtained directly from household surveys in both countries. For this comparison, the 20% difference in gross domestic product per capita in favour of the United States in 2000 is divided almost equally between labour productivity (-11%) and work intensity (-9%). As was the case with measure #1, this one also attributes more importance to differences in labour productivity than the estimate that our Canada/U.S. project yields (line 3). The differences with our reference measure originate mainly, in this case, with the absolute measures: the number of jobs and the working age population for the United States. Although, at first glance, Canada s Labour Force Survey (LFS) and the U.S. Current Population Survey appear to provide fully comparable estimates, a more detailed analysis of these two surveys reveals unsuspected and quite substantial differences due to data coverage. These differences are enough to compromise use of these surveys for direct comparisons of levels of jobs though not for hours worked per job when comparable methodology is applied to each survey. While both countries use similar questionnaires, the statistical agencies on opposite sides of the border do not have access to a similar method to calculate the survey frame. In Canada, the demographic weights of the Labour Force Survey are recalibrated every five years using a five-year census, while in the United States, this recalibration occurs only once every ten years. In addition, Canada s recalibration results in an historical revision of the LFS estimates to eliminate any break in the series. In contrast, in the United States, this exercise leads to significant breaks in the Current Population Survey (CPS) series, the most recent having occurred in 2000 and Added to this statistical problem is the much higher proportion of illegal immigrants in the United States, whom Bureau of Labor Statistics 2 analysts suspect are somewhat reluctant to respond to the CPS survey. However, legislation requires employers to report all of their 2. See Nardone et al., Examining the Discrepancy in Employment Growth between the CPS and the CES, FESAC, October The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

11 employees annually to unemployment insurance officials and this approach appears to provide a better estimate of illegal immigrants. In the United States, it is mandatory to have a social security number in order to obtain a job. It is the data from this file that are used as an annual benchmark for the employer survey (Current Employment Survey [CES]), which would explain why exercises to reconcile the two surveys indicate a substantial under-counting of jobs in the CPS compared to the CES. This problem was particularly evident between 1996 and Corrections made to the population estimate projection model by the U.S. Census Bureau have made it possible to narrow considerably the differences in job estimates between the two surveys since As a result, data from the U.S. household survey (CPS) frequently suffer from a problem of underestimating the levels of jobs and the working age population. It also suffers a problem since it only partially revises its series when benchmarking to the Census and this causes breaks in its historical series. These two problems make using job estimates from this survey inappropriate for Canada U.S. comparisons. Measure #3: Reference measures from the Canada/U.S. project for comparing levels Since the last historical revision of the National Accounts, the Canadian Productivity Accounts (CPA) have developed a measure of the number of hours worked that can be used to measure both the growth and level of labour productivity. This is why Canadian estimates of the number of hours worked and the number of jobs in measure #3 correspond to the estimates published by the CPA. In their project to compare Canada United States productivity levels, analysts with the Canadian Productivity Accounts selected their U.S. data sources to be comparable with the Canadian CPA data. For several years, the Bureau of Labor Statistics productivity program has also produced a level of employment that corresponds to the National Accounts concept, which covers the entire American economy and represents the most reliable level of employment that can be developed for that country. These are the estimates derived from their employer survey. However, there is a problem of comparability with respect to hours per job as described above. As part of the Canada/U.S. comparison project, analysts in the Canadian Productivity Accounts produced estimates of hours worked per job using the Current Population Survey and a similar methodology used for Canadian data to account for holiday bias. It is these estimates that were used to compare the sources of differences in the level of gross domestic product (GDP) per capita. Lastly, the population estimates used in this article are based on the concept of resident population. This concept is the one used in international GDP per capita comparisons. It is also important to note that it is the U.S. Census Bureau that produces these estimates using a revision procedure that avoids historical breaks. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

12 Although there are differences in the methodologies used by the two countries to produce hours worked estimates that enter into measures of the growth in productivity, as long as these differences remain constant, the accuracy of comparisons of growth rates in the two countries will not be greatly affected. However, these differences in methodology make comparisons of productivity levels more difficult and some care should be used in interpreting and using the data that have been used for comparisons of growth rates. In order to obtain more accurate estimates of productivity levels in Canada relative to the United States, effort is needed to harmonize data sources and methods. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

13 1. Introduction The Canadian Productivity Accounts (CPA) are one of the oldest productivity programs in the world. Since its creation in the early 1960s, this program has refined the concepts associated with the measurement of productivity growth and its related measures, has been integrated fully into the Canadian System of National Accounts, given an important place to international comparisons, and explored many productivity measurement issues. The program regularly provides comparisons of productivity growth in Canada and the United States. Although there are differences in the methodologies used by the two countries to measure productivity growth, as long as these differences remain constant, this type of comparison is meaningful. However, these differences make comparisons of productivity levels more difficult. In order to produce accurate estimates of the relative productivity levels in Canada as opposed to the United States, analysts must harmonize data sources and methods. Our previous study (Baldwin et al., 2005) constitutes the first attempt by the CPA to examine, in an exploratory way, the question of comparability of estimates of labour productivity in Canada and in the United States. Using analogous sources, concepts and methods to arrive at the most comparable measure of productivity levels possible, Baldwin et al., (2005) conclude that the Canada U.S. productivity level difference in 1999 was lower than normally described. The study by Baldwin, Maynard and Wong (2005, hereafter BMW) also uses the same methodology to compare gross domestic product (GDP) per capita in Canada and the United States and to decompose it into its constituent parts, namely labour productivity and work intensity. GDP GDP HRS EMP WAP = POP HRS EMP WAP POP { Labour productivity GDP HRS = HRS POP Work intensity (1) where GDP, POP, HRS, EMP and WAP represent, respectively, GDP expressed in comparable currencies using purchasing power parities, population, number of hours, number of jobs, and working-age population (the appropriate measures for these variables are discussed below). The HRS EMP ratios and are referred to, respectively, as average number of hours and the EMP WAP employment rate. Relying on the same sources, concepts and methods, BMW (2005) concluded that, during the period from 1994 to 2002, Canada s GDP per capita was approximately 80% that of the United States, largely because of the gap in the level of work intensity (hours worked per capita). The productivity level was approximately 90%. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

14 While BMW (2005) quantified the sources of the differences in the standard of living in Canada and the United States, they did not extensively explore the relative labour market performance in the two countries or the extent to which the latter could be reconciled with other labour market data published by the statistics bureaus of the two countries. The purpose of this study is to address those two concerns, while keeping in mind the need to illustrate the errors that could result from the use of series that are incompatible with one another. In addition, this study incorporates revisions and updated data up to 2005 that were presented for the first time in a study from Baldwin, Maynard and Wong (2005). 2. Comparison framework Although Canada and the United States are located on the same continent and their culture and institutions are similar, the statistical systems in the two countries rely on concepts and methods that are not always equivalent. There are two possible approaches that can be used to draw crosscountry comparisons using Canada/U.S. data: a) A mechanical approach is to use various labour market data published by the two statistical systems without considering the initial objective for which the series were established and whether series with similar titles are really comparable; b) A more time intensive approach is to compare sources, concepts and methods and to make modifications to the series of one or other country to reconcile differences. It was the latter approach that was adopted by Baldwin et al., (2005) and Baldwin, Maynard and Wong (2005), who made a considerable effort to ensure that the various components of the decomposition in Equation (1) were as comparable as possible in terms of concept and coverage. 3. Estimation of labour input for comparisons of relative levels of labour productivity in Canada and the United States There are a number of different sources that can be used to develop estimates of labour inputs for the purpose of comparing productivity levels in Canada and the United States. The suitability of particular sources depends on four factors: the extent to which they are consistent with the required concept, whether their coverage is appropriate, whether their methodology is comparable, and whether their accuracy is similar. 3.1 Criteria Concept An estimate of labour input for the purposes of analyzing productivity must allow for the measurement of the derived work effort that most accurately reflects the production of goods and services. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

15 Labour input can be measured by the number of persons employed or by the hours worked. Since workers do not work the same hours in every country, differences in effort are better reflected by the number of hours worked than by the number of persons employed. The 1993 System of National Accounts thus proposed hours worked as the preferred measure to be used with gross domestic product (GDP) for productivity estimates. Furthermore, the international definition of what constitutes work is based on time worked. The System of National Accounts (1993) uses a definition of hours worked that is consistent with the concept defined by the International Labour Office. The definition specifies that hours worked should include: - hours actually worked during normal periods of work; - time worked in addition to hours worked during normal periods of work, and generally paid at higher rates than the normal rate (overtime); - time spent at the place of work on work such as the preparation of the workplace, repairs and maintenance, preparation and cleaning of tools, and the preparation of receipts, time sheets and reports; - time spent at the place of work waiting or standing by for such reasons as a lack of supply of work, breakdown of machinery, or accidents, or time spent at the place of work during which no work is done but for which payment is made under a guaranteed employment contract; and - time corresponding to short periods of rest at the workplace, including tea and coffee breaks. And that hours actually worked should exclude: - hours paid for but not actually worked, such as paid annual leave, paid public holidays, paid sick leave; - meal breaks; and - time spent on travel to and from home and work. Any source of data on hours worked must include both hours worked during regular time and hours spent on work outside the regular schedule, whether paid or unpaid Coverage Estimation of labour inputs must correspond as closely as possible to the National Accounts production boundary, which serves to measure the production of goods and services. This applies to estimates of jobs, hours and population when gross domestic product per capita is calculated. Some sources for estimates of labour inputs do not cover all sectors. For example, agriculture is usually excluded from employer surveys. Some population aggregates also exclude a substantial number of individuals (i.e., those who live in institutions, such as long-term care facilities and penitentiaries or military personnel). Ideally, sources that provide only partial coverage need to be supplemented by data on the excluded part of the population. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

16 3.1.3 Accuracy or quality of estimates The accuracy of each estimate associated with a survey depends on both sampling error and non-sampling error. Sampling error will depend on the size of the sample and its design, while non-sampling error will depend on administrative practices, coverage problems and definitions. The quality of an estimate is partially dependent on its intended use. Some estimates may be highly appropriate for some uses and less so for others. For example, a particular source of labour data may be downward biased in terms of levels, while providing a good indication of the trend. Such a data source is appropriate for developing an estimate of labour growth used to derive estimates of labour productivity growth, but it would be inappropriate for estimating the level of labour productivity. In fact, as we note below, this issue is critical to the choice of a particular estimate of labour input for the United States and Canada if comparable estimates of the productivity levels in each country are to be produced Corroboration Discovering information that corroborates estimates of labour input is one way of evaluating the quality of such estimates. Alternative methods, albeit imperfect, can still be indicative of the appropriateness of the chosen estimate. In developing estimates of labour input for the purposes of comparing relative productivity levels in Canada and the United States, the Canadian Productivity Accounts focus on issues relating to coverage, concepts and accuracy. Harmonizing the definitions of labour to appropriate concepts and coverage is a priority for the Canadian Productivity Accounts because they are included in the Canadian System of National Accounts (SNA), which meet the 1993 SNA international standards with regards to GDP estimates. The method used to construct the appropriate measure of hours worked using different sources available also takes into account the accuracy and availability of each source. In addition, the Canadian Productivity Accounts examine other information to corroborate the results obtained. The chosen method is described in the pages that follow. We will begin with a description of the sources available. 3.2 Sources of labour inputs There are two main sources from which estimates of labour input for Canada and the United States can be produced, namely household surveys and employer surveys. The first collects information by asking members of selected households whether they are working and how much time they spend at work, whether paid or unpaid. The second asks employers directly for information on the number of people working at their businesses and the amount of time they work (normally their hours paid). Each of these surveys differs in terms of accuracy, although it is important to note that accuracy depends on the intended use for each source. What is appropriate for one use is not necessarily appropriate for another. We have already noted that what would be adequate for comparing the The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

17 employment growth rates in each country may not be adequate for comparing levels. Different series may provide essentially similar estimates of growth rates but different estimates of levels. It should be noted that producing accurate data in terms of levels is much more demanding in terms of statistical quality than what is necessary to provide a trend indicator. It is important to recognize that surveys are often developed to meet objectives that are different from those of a particular analyst especially those conducting cross-country comparisons. A household survey may be developed to provide information on short-term trends in the labour market but not necessarily to estimate the level of the employment population ratio. Moreover, a household survey does not necessarily constitute the best instrument for obtaining full coverage of all jobs in the economy, but may yield a more than adequate estimate of hours worked per job. In evaluating the extent to which a particular data source is appropriate for a particular use, an analyst needs to ask whether the respondent has the ability to provide the information requested. An equally important consideration is whether the statistical agency is able to deal with the estimation difficulties associated with a particular instrument used for data collection. Both household surveys and enterprise surveys encounter problems in obtaining the number of hours worked, which is required for measuring productivity. However, the problems and the solutions for dealing with them are different in each case Enterprise surveys Hours worked data from enterprise surveys contain several problems. The first is that firms often do not keep data on jobs that are not paid on an hourly basis. This includes white collar workers or the self-employed. It also includes workers with non-standard working arrangements. The latter make up a substantial part of the workforce. The Upjohn Institute reports that only 70% of workers are in jobs with standard work arrangements (Houseman, 1999). And of this group, only about 70% are hourly workers. This is becoming more of a problem in the service economy as contracts are often specified in terms of annual salaries with unspecified overtime commitments. A second problem occurs since enterprises can generally only report hours paid and not hours worked. And the size of unpaid hours worked has been increasing over the last two decades. In Canada, almost 9% of jobs report unpaid overtime, accounting for between 2% and 3% of total hours worked. 3 These problems have been dealt with in the United States in different ways. For example, the Bureau of Labor Statistics (BLS) supplements the hours worked estimates derived from an enterprise survey (the Current Employment Survey, or CES) for hourly workers with data on hours worked for salaried workers and self-employed workers taken from its household survey (the Current Population Survey, or CPS). Hours paid are transformed into hours worked with other information on how many hours worked are unpaid and on how many hours paid have not been worked (e.g., paid vacations, paid sick leave, etc.). 3. Special extractions from the 1998 Labour Force Survey. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

18 Enterprise surveys may also have problems obtaining data on hours worked from businesses if firms just do not keep track of hours worked data. As the work week becomes less standardized, firms have less of an incentive to keep hours worked as part of their management information systems. Indeed, Statistics Canada gave up asking questions about hours worked on its enterprise manufacturing surveys in the 1990s when the response rate to these questions fell well below 50% and resort to widespread imputations became extensive. Related to the problems that firms have in responding to surveys on hours worked is the extent to which outsourcing of payrolls has occurred. Many firms use third-party payroll firms and when hours worked data are supplied to the payroll firm, they are not kept permanently on computer files by these third-party payroll firms or by the originating firm since computer storage space is costly. If the enterprise survey is not done immediately after the payroll period, data are no longer available for hours worked even for hourly-paid workers Household surveys Household surveys have been developed with an extensive set of questions that permit statistical agencies to delve into the labour market status of household members, the type of work that they perform, and the number of hours including usual hours, and overtime hours, hours without remuneration and the reasons for time lost due to holidays, sickness, etc. When these surveys are conducted across different classes of workers (paid hourly, salaried, selfemployed), they generate estimates with good coverage. And since they ask for both paid and unpaid hours worked, they permit direct coverage of the definition of hours worked that meets international standards of work effort that can be compared to measures of output. While household surveys have the advantage over enterprise surveys in that they directly request information on the concepts required to meet international standards, household surveys do face various problems in providing error-free estimates of hours worked. First, in many households, the respondent will provide proxy answers for members of the household who are not present. And since respondents are asked for information on the previous week s experience, there may be a case of recall bias that is, respondents may not remember precisely the hours actually worked in the previous week. Survey methodologists in statistical agencies have devised ingenious methods to minimize these problems. The solution has been to design detailed questionnaires with special prompts as to unusual events in previous weeks, and to do follow-up surveys to gauge error rates. The result is a professional product in which most statistical agencies place great confidence. It is nevertheless the case that household surveys often need special editing because they are not continuous surveys and extrapolation of the results from the survey week to other weeks for the purposes of the Productivity Accounts requires recognition that holidays affect each week in a month differently. Household surveys may have problems with unusual events that occur during the reference week. The solution of the Canadian Productivity Accounts is to make detailed use of data on holidays and other events to provide corrected estimates for other weeks in a month. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

19 Enterprise surveys will not have problems with holidays that occur during the reference week if they report hours paid but to transform this estimate to number of hours worked to other periods not covered by the pay period requires transformations that are extremely complex Example: Differences between household surveys and enterprise surveys In Canada, the Survey of Employment, Payrolls and Hours, or SEPH (an employer survey in which hours data are collected from a sample of employers) collects monthly data on hours paid for workers paid by the hour and hours usually worked for workers with a fixed salary. Since the early 1990s, the percentage of employees paid by the hour has remained more or less constant, accounting for 52% of salaried employees in the SEPH universe in However, the percentage of workers with a fixed annual salary 4 decreased from 42% to 36% during the same period, while the category covering Other employees increased from 6% to 12% of jobs covered by the SEPH. The SEPH does not collect data on hours for these employees. The concept of hours paid and that of normal hours does not correspond to that of number of hours people work that is collected in the Canadian household survey (Labour Force Survey, or LFS) and that is required for the purpose of productivity measurement. 5 By definition, hours paid or normal hours worked includes paid hours absent from work for vacation, holidays, illness and other leave provided for in employment contracts. To calculate the hours people work, these components must therefore be estimated and then subtracted from hours paid. The concept of hours paid has a second problem: it includes paid overtime but not unpaid overtime. Transforming these concepts of hours paid into hours spent at work required by the System of National Accounts definition therefore requires adjustments using additional data. 6 Since 1997, the LFS has collected data on the two types of overtime separately. These data show that unpaid overtime is extremely significant, especially in the service industries. As a percentage of hours worked, the LFS indicated in 2002 that unpaid overtime accounted for approximately 2.4% of total hours worked, slightly more than the 1.8% represented by paid overtime. Not taking unpaid overtime into account thus results in significant underestimation of hours worked (see Maynard and Sunter, 2003). 4. Hours usually worked represent the regular work schedule of salaried employees. This variable would be 37.5 hours per week for federal employees, for example. 5. For a comparison, see Table A1 in Appendix See Table A3 for an attempt to reconcile the paid worker concept from the Canadian employer survey (Survey of Employment, Payrolls and Hours) with the hours worked concept from the Canadian household survey (Labour Force Survey). The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008

20 Table 1 Comparison of estimated aggregations of hours by job according to adjusted data from household surveys with those derived from employer surveys, annualized data, 2003 Canada United States Difference (Canada U.S.) A Adjusted household surveys 1, , B Employer surveys 1, , Difference (A B) not applicable Notes: Estimates for Canada are based on data from the Labour Force Survey and the Survey of Employment, Payrolls and Hours; for the United States, adjusted hours were compiled by the Canadian Productivity Accounts based on data from the Current Population Survey while the estimates from the employer survey correspond to hours per job estimated by the Bureau of Labor Statistics productivity program. Source: Statistics Canada, Canadian Productivity Accounts. In Table 1, we compare hours worked per job obtained from household surveys with those derived from employer surveys. For the United States, the estimates correspond to the estimate of hours worked taken from the Bureau of Labor Statistics productivity growth program. 7 The starting point for the Canadian estimates is data on hours paid for employees paid by the hour, including overtime, combined with the number of hours that reflect the regular work week of workers receiving a fixed annual salary as collected under the SEPH. To transform this data into hours worked, we deducted paid hours absent as determined by the LFS. Hours worked by workers not covered by the SEPH, such as those in agriculture, religious organizations and private households as well as all self-employed workers, also come from the LFS (see Table A2 in attachment). Table 1 shows that, for both countries, the data on hours worked derived from employer surveys are lower than those calculated using the data from household surveys. This underestimation is approximately 125 hours in Canada 8 and 130 in the United States. Hours worked derived from employer surveys are therefore not comparable to those obtained from household surveys, at least for these two countries. This table also suggests the average American works at least 100 hours more than the average Canadian (differences expressed in the last column of the table), regardless of whether the comparisons are derived using the household or the employer surveys Overall reliability of average hours worked derived from household surveys As argued above, household surveys are more likely to be able to provide the coverage of hours worked that corresponds to hours actually devoted to the production process than can enterprise surveys. The sign of the differences between the estimates produced by these types of surveys bears out our expectations. 7. These estimates are obtained by combining hours paid collected from the employer survey (Current Employment Survey) with Current Population Survey hours worked data to fill the employees categories and industries not covered by the CES. An annual compensation survey is also used to estimate the hours paid not worked due to holidays, vacation, etc. 8. For more detail on industry differences, see Appendix 1, Table A2. The Canadian Productivity Review Statistics Canada Catalogue no XIE, no. 008